Remote Sensing of Environment 115 (2011) 3080–3090

Contents lists available at ScienceDirect

Remote Sensing of Environment

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e

Identification and analysis of urban surface temperature patterns in Greater Athens,

Greece, using MODIS imagery
Iphigenia Keramitsoglou a,⁎, Chris T. Kiranoudis b, Giulio Ceriola c,
Qihao Weng d, Umamaheshwaran Rajasekar d
a
Institute for Space Applications and Remote Sensing, National Observatory of Athens, Metaxa & Vassileos Pavlou Str, GR 152 36, Palea Penteli, Athens, Greece
b
School of Chemical Engineering, National Technical University of Athens, Zografou Campus, GR 157 80, Athens, Greece
c
Planetek Italia S.r.l., via Massaua, 12-70132 Bari, Italy
d
Center for Urban and Environmental Change, Department of Earth & Environmental Systems, Indiana State University, Terre Haute, IN 47809, USA

a r t i c l e i n f o a b s t r a c t

Article history: Thermal infrared images are being acquired by satellites for more than two decades enabling studies of the
Received 28 December 2010 human-induced Urban Heat Island (UHI) phenomenon. As a result, the requirement of the scientific
Received in revised form 6 June 2011 community for fast and efficient methods for extracting and analyzing the thermal patterns from a vast
Accepted 15 June 2011
volume of acquired data has emerged. The present paper proposes an innovative object-based image analysis
Available online 22 July 2011
procedure to extract thermal patterns for the quantitative analysis of satellite-derived Land Surface
Keywords:
Temperature (LST) maps. The spatial and thermal attributes associated with these objects are then calculated
Land Surface Temperature and used for the analyses of the intensity, the position and the spatial extent of UHIs. A case study was
Urban Heat Island conducted in the Greater Athens Area, Greece. More than 3000 LST images of the area acquired by MODIS
Data mining sensor over a decade were analyzed. Three daytime hot-spots were identified and studied (Megara, Elefsina-
Object-based analysis Aspropyrgos and Mesogeia). They were all found to exhibit similar behavior, gradually increasing their
MODIS maximum temperature during the summer season and reaching their maxima in mid-July. The hot-spots'
thermal intensities compared to a suburban area were of 9–10 °C and were found to be highly correlated to
their areal extent. During the night-time, Athens center developed a typical UHI spatially coinciding with the
dense urban fabric. The nighttime maximum LST peaked (on average) at the end of July, two weeks later than
the daytime surface patterns. The mean spatial extent of UHI in Athens was 55.2 km 2, whilst its mean
intensity was 5.6 °C. The proposed automatic extraction process can be customized for other cities and
potentially used for comparison of LST patterns and UHI behavior between different cities.
© 2011 Elsevier Inc. All rights reserved.

1. Introduction
Knowledge of Land Surface Temperature (LST) and its temporal
and spatial variations within a city environment is of prime
importance to the study of urban climate and human–environment
interactions (Hung et al., 2006; Stathopoulou & Cartalis, 2009; Weng,
2009; Weng & Quattrochi, 2006). Studies on surface Urban Heat
Islands (SUHI) and LST distribution within a city have been carried out
mostly by using ~ 1 km spatial resolution satellite sensors, such as
AVHRR (Advanced Very High Resolution Radiometer) or MODIS
(Moderate Resolution Imaging Spectroradiometer) (Balling & Brazel,
1988; Caselles et al., 1991; Dousset & Gourmelon, 2003; Gallo & Owen,
1998; Gallo et al., 1993a,b; Hung et al., 2006; Kidder & Wu, 1987; Lee,
1993; Rajasekar & Weng, 2009; Roth et al., 1989; Stathopoulou et al.,
⁎ Corresponding author. Tel.: + 30 210 810 9167; fax: + 30 210 613 8343.
E-mail addresses: ik@noa.gr (I. Keramitsoglou),
kyr@chemeng.ntua.gr (C.T. Kiranoudis), ceriola@planetek.it (G. Ceriola),
qweng@indstate.edu (Q. Weng), rumahesh45@gmail.com (U. Rajasekar).
2004; Streutker, 2002). The LST of urban surfaces correspond closely
to the distribution of land use and land cover (LULC) characteristics
(Lo et al., 1997; Weng, 2001, 2003; Weng et al., 2004). Moderate and
high spatial resolution thermal infrared imagery, such as those from
Landsat TM/ETM+, ASTER (Advanced Spaceborne Thermal Emission
and Reflection Radiometer), and airborne ATLAS data, have been
extensively employed to study intra-urban temperature variations
and to relate them to surface cover characteristics (Aniello et al., 1995;
Carnahan & Larson, 1990; Kim, 1992; Nichol, 1994, 1996, 1998;
Stathopoulou & Cartalis, 2007a; Weng et al., 2006, 2008; Xiao et al.,
2008). However, the utilization of high and moderate spatial
resolution images in urban climate studies is somewhat restricted
because of the limited available data (especially night-time images)
and low temporal resolution (Stathopoulou & Cartalis, 2009). To
bridge the gap, Pu et al. (2006) assessed data from thermal sensors
with different spatial resolutions, including ASTER and MODIS, and
concluded that the MODIS thermal sensor can be used for the synoptic
overview of an urban area, and that the ASTER sensor can be used for a
more accurate/detailed determination of thermal patterns and LST.

0034-4257/$ – see front matter © 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2011.06.014
 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090
Despite the large number of publications on LST and Urban Heat
Islands (UHI) studies using satellite and airborne sensors, Voogt and Oke
(2003) criticized that thermal remote sensing of urban areas had been
slowly progressed due largely to qualitative description of thermal
patterns and simple correlations between LST and LULC types.
Furthermore, Xiao et al. (2008) noticed that little research has been
done on the statistical relationship between LST and non-biophysical
factors. Recent review articles by Weng (2009) and Stathopoulou and
Cartalis (2007a), as well as the editorial of the special issue of Remote
Sensing of Environment “Thermal remote sensing of urban areas” by
Weng and Quattrochi (2006) examined research progresses in this
particular field, identified existing problems, and speculated on its
future development. A key issue in the application of TIR remote sensing
data in urban climate studies is how to use LST measurements at the
micro-scale to characterize and quantify surface temperature patterns
observed at the meso-scale (Weng, 2009). This observation further
implies the need for developing automatic extraction methods and
customized software to handle efficiently large volumes of satellite data
used for spatial–temporal analysis. Streutker (2002, 2003) used AVHRR
data to quantify the UHI of Houston, Texas, as a continuously varying
two-dimensional Gaussian surface superimposed on a planer rural
background, and derive the UHI parameters of magnitude (i.e.,
intensity), spatial extent, orientation, and central location. Hung et al.
(2006) adopted the Gaussian method proposed by Streutker (2002) to
measure the spatial extents and magnitudes of the SUHIs for eight
mega-cities in Asia. They used both daytime and night-time MODIS data
acquired over the 2001–2003 period to produce surface temperature
maps for the eight cities at 1 km spatial resolution, and revealed their
diurnal and seasonal patterns. Hung et al. (2006) studied the
relationship between heat islands and surface properties as well as
the correlation between SUHI magnitude and city population. Rajasekar
and Weng (2009) applied a non-parametric model by using fast Fourier
transformation (FFT) to MODIS imagery for characterization of the UHI
over space, so as to derive UHI magnitude and other parameters. Weng
(2009) in his exhaustive review on thermal infrared remote sensing for
urban studies pointed out that in spite of these advances, new methods
for estimation of UHI parameters from multi-temporal and multi-
location thermal infrared imagery are still needed given the increased
interest in urban climate community to use remote sensing data.
The main objective of our study is to develop a fully automated
extraction method for the analysis of surface temperature patterns from
a large number of LST images. Specifically, we intend to obtain a fast
procedure to analyze hundreds of LST maps; to identify and extract
surface temperature patterns (hot-spots); to characterize the hot-spots
(in terms of spatial extent and intensity); and to investigate any
correlations between spatial and thermal attributes, based on a case
study of the Greater Athens Area in Greece. The methodology developed
comprises consequent modules of data pre-processing, feature extrac-
tion, database generation, and data analysis. It must be pointed out that
Weng (2009) referred to the confusion that exists in published literature
between UHIs and ‘surface temperature patterns’, and commented that
the latter is more meaningful; thus, we have adopted the same
terminology in the present paper. We have also used the term ‘hot-
spots’ to help the reader visualize the increased LST of the patterns under
discussion compared to the cooler suburban/rural areas.
2. Study area and data sets
2.1. Athens Greater Area, Greece
The study area is the Athens Greater Area, the capital and largest
city of Greece. According to the recent census paper of Eurostat
(http://epp.eurostat.ec.europa.eu/), the Athens Larger Urban Zone
(LUZ) is the 8th most populated LUZ in the European Union with a
population of about 4,000,000. The area lies at the southeasternmost
edge of the Greek mainland (Fig. 1). Athens sprawls across a central
3081
basin bound by four mountains and the Saronic Gulf in the
southwest. The basin is bisected by a series of small hills. These
specific topographic characteristics make Athens an example of a
coastal city located in very complex terrain, where sea-breeze and
heat-island circulations interact (Dandou et al., 2009) making it
especially interesting for UHI and thermal pattern studies. In terms
of climate, Athens enjoys a prolonged warm and dry period during
the year with July and August being the hottest and driest months.
The normal value of the summer (from June to August) daily
maximum temperature at Athens is 31.6 °C (with reference to
1961–1990 period; Founda & Giannakopoulos, 2009). In summer-
time the city suffers from heat wave events. Particularly in June 2007
a severe heat wave lasted for seven days reaching temperatures as
high as 46 °C.
In the daytime the LST patterns of Athens Greater Area yield
the presence of two dominant and one weaker hot-spots as
shown in Fig. 1. The two dominant hot spots occur at the
Municipalities of Megara (denoted by rectangular ‘1’ in the figure)
and Elefsina and Aspropyrgos (‘2’) whilst the weaker one occurs
at the area of Mesogeia (‘3’). During night-time a typical UHI is
developed related clearly to the urban and industrial zone city
center (‘4’). These will be exhaustively discussed and analyzed in
Section 4.
2.2. Data sets
A nine-year time series of MODIS summer images was analyzed.
Specifically, 1085 daytime MODIS-Terra LST images and 1956
nighttime MODIS-Terra and Aqua LST images were produced and
archived as part of the requirements of Urban Heat Islands and Urban
Thermography project (21913/08/I-LG) funded by the European
Space Agency. In total more than 3000 raster LST maps of 1 km
spatial resolution were tested and processed using the algorithm
described in this paper.
The original MODIS data used in the present study were procured
at no charge from the Warehouse Inventory Search Tool (WIST web
site) repository. Firstly, the MODIS onboard Terra satellite was
considered and all the available scenes covering the area of interest
were selected. The subset of the daytime scenes with a central
viewing angle larger than 45° was identified and removed, due to the
well-documented directional dependence of the LST (Lagouarde &
Irvine, 2008).
The selected TERRA images consisted of one day- and one night-
time scenes, although a number of days (about 2 days every 6) had
no available datasets due to the viewing angle constraint. Conse-
quently, MODIS Aqua images were selected so as to ensure that at
least one daytime scene is available per day and 1–2 night-time
scenes are also available considering that the UHI effect is more
evident during the night. At that stage no cloud coverage limitation
was set; nevertheless, this was taken into consideration further in
the processing chain.
3. Methodology
Initially the MODIS images were processed and the corresponding
LST maps were produced. Thermal pattern analysis was implemented
through a series of computational procedures involving consequent
modules of data pre-processing, feature extraction and data mining.
Hot-spots were then analyzed to reveal spatial–temporal trends, as
schematically shown in Fig. 2. The complete algorithmic procedure for
the thermal pattern analysis was coded in C# .NET, which includes gap
filling, filtering, global thresholding, generation of pixel regions and
extracting features from regions. The sequence of procedures is
described next.
3082 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090

Fig. 1. A typical LST map of Greater Athens Area, Greece, during summer. Numbers 1 to 4 denote the four characteristic surface temperature patterns (hot-spots) of the study area,
namely 1: Megara, 2: Elefsina-Aspropyrgos 3: Mesogeia and 4: City Center. A representative subarea of each hot-spot is also provided. (Photo sources: Dounis airfield—I. Saridakis,
Google Earth™ mapping service, www.airphotos.gr and Microsoft Bing Maps Platform data, respectively).

3.1. LST computation
The methodology implemented for the calculation of the LST maps
is based on the Split Window Technique (SWT) from Jiménez-Muñoz
and Sobrino (2008). It has been applied to a number of thermal
infrared sensors available today (e.g. ERS ATSR and ATSR2, ENVISAT
AATSR, MODIS and NOAA AVHRR) by means of the calibration of some
coefficients. The general SWT formula adopted is:
   2
TS = Ti + c1 Ti −Tj + c2 Ti −Tj + c0
+ ðc3 + c4 PW Þð1−εÞ + ðc5 + c6 PW ÞΔε
where Ti and Tj are the at-sensor brightness temperatures of the split
window bands i and j (in Kelvin), ε is the mean emissivity ε = 0.5(εi +
εj), Δε is the emissivity difference, Δε = εi − εj, PW is the total
atmospheric water vapor content (in g cm − 2), and c0 to c6 are the
SW coefficients specific for each sensor, taken from a predetermined
table calculated by Jiménez-Muñoz and Sobrino (2008) with
MODTRAN simulations. In the case of the MODIS sensors i and j are
bands 31 and 32, respectively. The remaining unknown parameters
PW and ε were retrieved using various techniques which have the
advantage of needing only the MODIS data itself.
The retrieving of water vapor concentration for daytime images was
based on the ratio technique (Sobrino et al., 2003) which uses the MODIS
bands in the near infrared spectrum that are most sensitive to PW:
ð1Þ
PW = 0:192PW17 + 0:453PW18 + 0:355PW19 ð2Þ
where
PW17 = 26:314−54:434G17 + 28:449G217 ; G17 = L17 =L2 ð3Þ
2
PW18 = 5:012−23:017G18 + 27:884G18 ; G18 = L18 =L2
2
PW19 = 9:446−26:887G19 + 19:914G19 ; G19 = L19 =L2

Fig. 2. Schematic representation of consecutive modules for the analysis of surface temperature patterns.
 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090
Li are the radiance measurements acquired by MODIS band i (i = 2, 17,
18, and 19).
For the night-time images an equation derived from a study of
Sobrino et al. (2001) on AVHRR using the Linear Split-Window
Relationship (LSWR) was adopted, assuming that bands 4 and 5 of
AVHRR are equivalent to bands 31 and 32 of MODIS, respectively:
PW = aðT31 −T32 Þ + b
where T31 and T32 are the brightness temperatures derived from
MODIS bands 31 and 32, respectively. For different emissivities, the
slope a and intercept b of Eq. (4) are provided in Sobrino et al. (1999).
In particular, for:
2 −1 −2
• ε = 0:96 ðΔε = 0:01Þ; α = 2:10 cm K and b = 1:93 gcm
2 −1 −2
• ε = 0:98 ðΔε = 0:00Þ; α = 2:21 cm K and b = 1:23 gcm
2 −1 −2
• ε = 1:00 ðΔε = 0:00 Þ; α = 2:22 cm K and b = 1:004 gcm
All a and b values refer to simulations presented in Sobrino et al.
(1999) assuming a negative gradient (between 0 and 5 K) between
LST and the temperature of the first boundary layer of the
atmosphere.
Finally, the emissivities for bands 31 and 32 were calculated with
a simplification of the Normalized Difference Vegetation Index
(NDVI) Thresholds Method (NDVI THM) introduced by Sobrino et al.
(2008). The NDVI THM method uses certain NDVI values (thresholds)
to distinguish between soil pixels (NDVI b NDVIS), pixels of full
vegetation (NDVIN NDVIV) and mixed pixels (NDVIS ≤ NDVI ≤ NDVIV),
assuming the typical threshold values of NDVIV = 0.5 and NDVIS = 0.2.
For MODIS
• if NDVI b 0:2 then ε31 = 0:984−0:088ρ1 ; ε32 = 0:982−0:028ρ1
• if 0:2 ≤ NDVI ≤ 0:5 then ε31 = 0:974 + 0:015Pv ; ε32 = 0:968 + 0:021Pv
• if NDVI N 0:5 then ε31 = ε32 = 0:99
where ρ1 is the reflectivity in band 1 and PV is the proportion of
vegetation in a mixed pixel, also referred to as fractional vegetation
cover, and is given by
 2
NDVI−NDVIS
PV =
NDVIV −NDVIS
3.2. LST maps pre-processing
The input data were the MODIS LST products, as described above.
The LST maps were in Hierarchical Data Format, commonly abbrevi-
ated HDF. The data pre-processing included masking out not valid
pixels based on a corresponding flag to assure that only cloud free, not
obliquely viewed (N40°) land pixels would be used further in the
analysis. Once the valid images and pixels were selected, the problem
of missing (flagged) pixels had to be tackled by applying an averaging
technique on invalid pixels, where their LST value was actually
substituted by the average LST value of their adjacent valid pixels. This
procedure, based on a kernel convolution method suggested by
Rajasekar and Weng (2009), can be applied in an iterative manner
deploying subsequent passes through the invalid pixels of the image.
In our case, a single pass was selected as the procedure's critical step
since complete restoration would probably produce erroneous
patterns within the image. The procedure adopted was necessary in
order to facilitate the process of automated information extraction
from a large set of images. The resulting image was subsequently
3083
filtered using a similar averaging technique as the one presented in
the gap filling problem. Filtering was used for smoothing pixel LST so
that extreme values were appropriately throttled.
The next step of the pre-processing was the extraction of hot-spot
objects through global image thresholding. In the case of global
thresholding, the window addressed was the entire image. All pixels
whose LST was higher than the suburban LST plus a predefined
ð4Þ threshold value (let us say 6 °C) were registered as potential hot
pixels. Therefore, the definition of the suburban area, which would
serve as the reference area for the analysis, was an important issue.
The area selected was the municipality of Thrakomakedones in
Attica. The municipality is located at the foot of Parnitha Mt. It is a
sparsely built up municipality near a forested area. For each
individual image, the procedure evaluated the mean LST for the
pixels corresponding to this municipality and hereafter we will refer
to this temperature as RLST (Reference Land Surface Temperature). It
is stressed that each LST image had its own RLST value. Following the
calculation of the reference LST, each pixel LST of the image was
compared to RLST.
3.3. Feature extraction
An object-based image analysis procedure was developed to
extract hot-spots. According to this procedure, hot pixel image
groups were extracted using a global thresholding algorithmic
procedure. Separation of hot pixel groups was automatically
performed through appropriate segmenting the initial image by
partitioning the pixels whose LST value was above a threshold
value plus the average LST of reference area pixels. For these pixels,
segmentation was performed by determining the k-groups (Pal &
Pal, 1993). This type of groups was characterized by a certain
proximity property for the pixels of the group, that is to say, each
pixel in the group had at least one pixel neighbor that was close to
the former by a distance of k pixels in any direction. This distance
metric adopted was actually the norm of the digital space
processed (image). In this way, depending on the value of k,
smaller groups could be merged to bigger ones. In our system, the k
pixel distance parameter was set to one (each pixel had at least
one neighbor at a distance of one pixel). An algorithm of linear
computational complexity was used to separate groups obtained.
According to this algorithm each pixel was checked for its
neighbors, which in turn were checked for theirs in a recursive
way. The grouped pixels were then treated as different regions
ð5Þ
(objects). Following that, several features related to these objects,
which represent the city hot-spots were extracted.
3.4. Data mining
The hot-spots represented as objects underwent a series of
procedures to extract the necessary information for the area. As a
first step a procedure counted the number of pixels that belonged to
each object. If the object area was smaller than 10 km 2 (i.e., number of
pixels was less than 10), then the object was not further studied. Only
the objects whose weighted centroid belonged to a predefined area of
interest were taken into consideration. Four areas of interest were
defined, already mentioned in Section 2.1. The results of the above
mentioned procedure are illustrated in Fig. 3.
Once the objects of interest were extracted a number of
parameters per object were calculated and appropriately stored in a
database. These parameters included the satellite sensor as well as
temporal, spatial (weighted centroid location, extent in km 2) and
thermal information (e.g. minimum, maximum LST). In total, 1237
daytime hot-spots were identified with a 6 °C threshold value and 358
night-time hot-spots with a 4 °C threshold.
3084 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090

Fig. 3. Schematic representation of the data mining concept. The areas warmer than a reference LST are identified, extracted and subsequently treated as objects. (a) Typical daytime
LST map with three hot-spots. The hot-spots are extracted and the relevant spatial and thermal characteristics of the objects are calculated and stored in a database, and (b) Typical
night-time LST map representing the UHI of Athens city.
 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090
Table 1
Day- and night-time hot-spots of the Athens Greater Area.
Hot- Hot-spot Municipalities
spot characterization
1 Dominant, Daytime Megara
2 Dominant, Daytime South part of Aspropyrgos and Elefsina
3 Weaker, Daytime At and around Koropion, Paiania, Spata-Loutsas (Mesogeia)
4 Dominant, Night-time City of Athens (Athinai), Ilion, Ag. Anargyroi, Nea Halkidona,
Peristerion, Egaleo, Kamatero, Nikaia, Keratsinion, Tavros, Ag.
Ioanni Renti, Peiraiefs, Moshaton, Kallithea, N. Smyrni, Drapetsona
4. Results
4.1. General thermal pattern in the Greater Athens
Four areas of interest can be defined in the majority of the LST
maps for the Athens Greater Area. In particular there are three
daytime hot-spots (Hot-spot No. 1, 2, 3; Fig. 1 and Table 1) and one
night-time (Hot-spot 4; Fig. 1 and Table 1). On the daytime images
(most of which were acquired before 12:00 Local time), surface
temperature spatial patterns revealed that the center of Athens did
not exhibit the highest LST; this is referred to as urban heat sink or
negative heat island. It can be simply considered as a brief stage in the
development of the UHI that occurs during the morning and midday
period due to differences in the urban–rural warming rates (Oke,
1987). This behavior is well known for Athens, as already reported in
Stathopoulou and Cartalis (2007b). Therefore, in the daytime the LST
patterns of Athens Greater Area yielded the presence of two dominant
and one weaker hot-spots, as presented in Fig. 1 and described in
Table 1.
The typical daytime pattern in 3D can be also observed in Fig. 4(a).
In this figure the LST images were plotted in a wire frame model to
appreciate visually the thermal variations as ‘height’ (z-axis) and
color. Visual analysis of the hot-spots (especially the two dominant
ones, hot-spots Nos. 1 and 2; Table 1) showed that there was a LST
discrepancy of at least 4–6 °C from the suburban areas. They appeared
in almost every morning image of June–August; the phenomenon was
weaker in May and September. These hot-spots were clearly
identifiable by photo-interpretation of satellite images on all spatial
resolutions (from Landsat TM to MSG-Seviri, not shown here). It
should be noted that the time of MODIS acquisitions allowed studying
of the phenomenon at certain time (around 09:00 UTC; midday 12:00
local summer time). Another interesting feature of Fig. 4(a) is the
presence of cooler pixels along the coastline, which may be partially
caused by sea breeze. In general, the thermal environment of Athens
during daytime depends on the combined influence of the area
topography and surface cover characteristics. The open plain of
Mesogeia (Hot-spot No. 3, Table 1 and Fig. 1) is mainly covered with
sparse low vegetation (particularly olive trees and vineyards) and
bare soil (Athens International Airport “El. Venizelos” is also located at
this area) whilst Aspropyrgos and Elefsina (Hot-spot No. 2 of Fig. 1)
are mainly industrial zones. These plains become warm faster than
densely built-up urban areas which are extensively covered by
materials of high thermal inertia such as concrete and asphalt. The
influence of topography is also evident on the thermal pattern of the
study area, as higher altitudes exhibit lower LST.
During the night-time, the thermal pattern of Athens was inverted
as higher surface temperatures were related with the residential
urban zones rather than the different urban use zones and rural areas.
There was one dominant hot-spot affecting a number of central
municipalities, as described in Table 1. A 3D representation of a typical
night-time LST image is shown in Fig. 4(b). At that time of day, cooling
or warming of a surface is determined by its thermal characteristics.
3085
Corine land Corine land
cover class cover code
Agricultural land 2.2.3, 2.4.2, 2.3.1
Industrial zone discontinuous urban fabric, 1.2.1, 1.1.2, 1.2.4,
airport and agricultural land 2.4.2
Agricultural land, Airport 2.4.2, 2.2.1, 1.2.4
Urban areas (mainly continuous urban Mainly
fabric) 1.1.1
Thus, during night hours, the continuous urban fabric is a few degrees
warmer than rural areas, due to the lower thermal inertia of the soil
compared to concrete. In contrast, the morning hot-spots of the city
have faded out in the night and appear to be cooler than the
continuous urban fabric owning to the fact that in industrial and
agricultural areas usually extended open spaces of bare soil cover
most of the area. The presence of a heat wave increases LST and makes
less definable the hot-spots.
4.2. Time series of maximum LST
The high volume of data allowed the extraction of statistics and the
investigation of possible correlations and trends. In this paper we
present the results of the analysis including the behavior of the
maximum LST of the different hot-spots over time, the frequency of
occurrence of the thermal intensity (discrepancy between hot-spot
maximum and RLST), and the correlation between the thermal
intensity and the size of the hot-spots.
The maximum LST of the hot-spots (hereafter referred to as
maxLST) was analyzed by plotting the maxLST acquired between
08:55 and 10:00 UTC against date (1st May to 30th September) for
every year separately and fitting a quadratic polynomial to the data.
The reason behind the selection of a certain time window was to
eliminate the diurnal variation signal from the analysis. In any case
this specific window was the one with most of the daytime MODIS
acquisitions ensuring minimum loss of data. The results per year and
hot spot are presented in Table 2, whilst for all years are shown in
Fig. 5 (day = 0 is the 1st May and day = 152 is the 30th September;
the year was irrelevant). The plot clearly shows a consistent seasonal
variation with increasing maxLST from the beginning until approx-
imately middle of the season and then the gradual decrease of
temperature until the end of the season (end of September). Fig. 5(a)
shows the described seasonal trend of Megara Hot-spot (No. 1) with a
maximum of 48.6 °C on July 14, after when the maxLST starts
decreasing. The same behavior was observed at Elefsina-Aspropyrgos
hot-spot (No. 2); in this case the maximum was observed on July 16
and was of the same value. The third slightly weaker hot-spot of
Mesogeia (No. 3) was consistently cooler (between 1 and 3 °C;
Table 2) compared to the other two dominant hot-spots. As a result
the maximum of the overall seasonal trend (Fig. 5(a)) was 2 °C cooler
and was observed on July 15.
For the night-time images, a 4 °C threshold was selected for the
city center, as it was more realistic and increased the number of
objects and subsequently statistical samples. The maxLST for the city
center (Fig. 5(b)) follows a similar seasonal pattern. As expected the
maxLST was considerably lower than the morning hot-spots. The
quadratic line fitted to all the available maxLST night-time data
versus day of the season peaked on July 29, two weeks later than the
daytime surface patterns. The same behavior was observed for all
years (Table 2). By definition, this was the UHI pattern of Athens
city.
3086 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090

Fig. 4. Typical 3D LST patterns in Athens Greater area using MODIS. (a) Daytime hot-spots 1, 2 and 3. For a fly though a typical daytime LST 3D distribution of Athens Greater Area
please visit: http://195.251.203.191/video/daytime.html; (b) Night-time hot-spot 4. A fly though a typical night-can be seen at http://195.251.203.191/video/nighttime.html.

4.3. Hot-spots intensity and extent intensity shows the LST difference between the hot-spot and the
suburban area. As before, the three daytime hot-spots and the one
By thermal intensity we refer to the discrepancy between the night-time hot-spot of the urban center were studied. Fig. 6 shows
maxLST of a hot-spot and the RLST. In other words, thermal the distribution of the Intensity occurrences per hot-spot by means
of histograms. A Gaussian curve was fitted on the original data, for
Table 2 better statistical description of the results. The daytime results
Fitted maximum LST of the four hot-spots of Athens Greater Area per year with the for Megara are presented in Fig. 6(a) and for Elefsina-Aspropyrgos
corresponding date when it occurred. in Fig. 6(b). The Intensity occurrences were similar for both areas
Year Hot-spot
and both distributions peaked at 10 °C. The corresponding his-
togram for Mesogeia hot-spot is shown in Fig. 6(c). The Gaussian
1 (Daytime) 2 (Daytime) 3 (Daytime) 4 (Night-time)
curve in this case peaked at 9 °C. Fig. 6(d) gives the distribution
Date Max LST Date Max LST Date Max LST Date Max LST for the night-time measurements at the City Center, which—by
2000 7-Jul 50.2 9-Jul 50.3 10-Jul 47.9 19-Jul 26.4 definition—was the UHI magnitude. The fitted curve peaked at
2001 13-Jul 49.5 15-Jul 49.0 12-Jul 46.3 1-Aug 27.8 5.7 °C at night-time, decreasing faster than the corresponding day-
2002 11-Jul 47.9 16-Jul 49.1 16-Jul 46.8 30-Jul 26.1 time curves.
2003 17-Jul 48.0 18-Jul 48.2 19-Jul 45.5 29-Jul 25.8
2004 16-Jul 48.7 18-Jul 49.4 15-Jul 47.0 2-Aug 26.5
Whilst the intensity was the difference between maxLST and RLST,
2005 17-Jul 48.3 16-Jul 48.0 18-Jul 45.9 29-Jul 25.9 tDiff was defined as the difference between maximum and minimum
2006 13-Jul 46.6 12-Jul 47.4 12-Jul 45.3 24-Jul 26.1 LST within the same surface temperature pattern, thus giving a
2007 12-Jul 49.2 14-Jul 48.4 15-Jul 46.9 1-Aug 27.4 measure of the thermal differences inside an affected area. The mean,
2008 12-Jul 50.4 17-Jul 49.1 10-Jul 47.4 27-Jul 26.2
the maximum and the absolute (standard) deviation (ADEV) of the
 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090 3087

Fig. 5. Maximum LST acquired at (a) the three daytime hot-spots between 08:55 and 10:00 UTC during the summer months (May–September) from 2000 to 2009 as a function of day
of season, and (b) Athens City Center hot-spot (No. 4) at night-time for the same period.

hot-spots surface extent, intensity and tDiff were calculated. Table 3
presents the results.
In terms of the surface extent the weaker morning hot-spot of
Mesogeia is the largest, the second being the dominant hot-spot of
Megara and then of Elefsina, with an average surface of 112.0, 95.2
and 83.5 km 2, respectively. The dominant hot-spot of Elefsina,
however, is the least variable one (the ratio ADEV/MEAN of the fitted
values is 0.33 compared to 0.61 for the extended and quite variable
hot-spot of Mesogeia). In spite of differences in surface extent, all
three morning hot-spots are consistent in mean thermal intensity,
which varied between 9 and 10 °C with similar statistical spread
(ADEV) of the values of less than 1.2 °C. It is interesting to note that
the maximum recorded thermal intensity is above 17 °C for all three
morning hot-spots and even reached 20.6 °C in Elefsina at 09:40 UTC
of 2007-05-13. The mean thermal difference (tDiff) within the
morning hot-spots ranges from 3 to 4 °C, although a maximum
difference of 14.6 °C inside the hot-spot was recorded in Elefsina on
the same satellite acquisition. The standard deviation (ADEV) is
between 1.0 and 1.2 °C.
The investigation of the night hot-spot behavior reveals the
characteristics of the UHI phenomenon of the center of Athens. The
mean surface extent of UHI pattern is 55.2 km 2 using the original
values and 3.6 km 2 larger using the gaussian fitted values. The reader
is reminded that the night-time patterns are extracted using a 4 °C
threshold difference between the hot-spot and the reference
suburban area. When increasing the threshold the spatial extent is
smaller; for instance for 6 °C threshold the mean is 45.8 km 2. The
mean UHI intensity is of 5.6 °C, also shown in Fig. 6(d). A maximum
intensity of 12.3 °C was recorded on 2004-09-11 at 20:30 UTC. The
mean thermal difference within the UHI pattern (tDiff) is lower than
the morning patterns and is equal to 1.6 °C. However, tDiff reached
8.3 °C on the same image.
In the previous paragraphs we have discussed the statistics of the
extent of the hot-spots, an interesting parameter as it reveals the area
affected by the higher temperatures. Streutker (2002) investigated
whether the spatial extent of the heat island was related to the thermal
intensity (referred to as UHI magnitude) and to the surrounding rural
temperature. This was found not to be the case in his research, as none of
these temperature measurements showed significant correlation to the
spatial extent. When we investigated similar correlations of the daytime
hot-spots of Greater Athens Area, we found that their hot-spot extent
was not correlated directly with a thermal feature such as the maxLST or
minLST. Nevertheless, our analysis showed that the spatial extent of a
hot-spot is highly correlated with its intensity. This is true for all three
morning hot-spots whose correlation coefficient is above 0.75, yet not
true for the UHI night-time city center phenomenon. In particular, the
correlation coefficients for the hot-spots of Megara and Elefsina-
Aspropyrgos are 0.79, whilst for Mesogeia is 0.76 (Fig. 7). This
correlation is reported for the first time for Greater Athens, since
published works has been mostly focused at the city of Athens and its
3088 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090

Fig. 6. Histogram representing the occurrences of certain thermal intensities (maxLST-RLST) (a) at Megara dominant hot-spot during daytime, (b) at Elefsina-Aspropyrgos dominant
hot-spot during daytime, (c) at Mesogeia weaker hot-spot during daytime, and (d) at Athens City Center during night-time. The latter is the UHI intensity.

immediate vicinity (e.g. Livada et al., 2002; Santamouris et al., 2007;
Stathopoulou & Cartalis, 2007a).
In order to explain the observed correlation, one has to consider
that LSTs estimated by satellite sensors are the spatial patterns of
upwelling thermal radiance received by the remote sensors. A hot-
spot is thus the aggregated result of warmer pixels. In general LSTs are
higher and more variable than concurrent air temperatures due to the
complexity of the surface types in urban environments and variations
in urban topography (e.g. Nichol, 1996; Streutker, 2002) and they are
more easily related to surface conditions themselves (Imhoff et al.,
2010; Nichol & Wong, 2005; Owen et al., 1998; Voogt & Oke, 2003).
Table 3
Statistics related to the characteristics of the three day- and one night-time hot-spots of
Athens Greater Area as calculated from MODIS data acquired during the summer months
(May–September) of years 2000–2009. Fitted values refer to a quadratic polynomial.
Hot-spot Statistical Surface extent Intensity (°C) tDiff (°C)
parameter (km2)
The observed daytime correlations presented here are strongly
related to the differential heating rate (Oke, 1982; Voogt & Oke,
2003) between open spaces such as bare soil and agricultural land
(like in Megara and Mesogeia hot-spots) compared to sparsely built-
up vegetated areas as our reference area. Land covered by vegetation
has a slower heating rate than bare soil or agricultural land and
reaches lower LST. In this work we have defined the thermal intensity
as the discrepancy between hot-spot maximum and RLST, and the
spatial extent of a hot spot as the area covered by the aggregated
cluster of pixels whose LST is higher than the RLST plus a predefined
threshold value. Let us now assume solar heating which will result
in LST rise. The rise will be higher in the area of the hot-spot than
the reference area due to differential heating rate. This in turn will
result in a two-fold effect: the thermal intensity will increase and
the cluster of warm pixels that will be extracted as thermal pattern
will be larger. It is worth noting that a hot-spot would potentially
increase in size until a topographic feature, a densely vegetated area
or a large water body imposes an abrupt change in the LST distribu-
tion (Oke, 1982).

Data Fitted Data Fitted Data Fitted

5. Conclusions

1 (Daytime) MEAN 95.2 97.4 10.1 10.2 4.1 4.2

MAX 561.0 330.5 17.9 14.6 11.8 8.5
This paper introduced an object-based image analysis procedure to
ADEV 47.3 36.2 1.6 1.2 1.6 1.2 extract hot-spots from a large volume of satellite thermal infrared
2 (Daytime) MEAN 83.5 86.9 9.8 9.8 3.7 3.8 image dataset. A series of attributes was calculated for each hot-spot
MAX 362.0 251.7 20.6 15.6 14.6 9.6 pattern, i.e., the aeria extent and thermal features, allowing the study
ADEV 40.8 29.1 1.5 1.0 1.5 1.0
of the spatial behavior of thermal patterns with time (over a decade
3 (Daytime) MEAN 112.0 129.0 9.0 9.1 3.0 3.1
MAX 667.0 474.2 17.3 12.7 11.3 6.7 or/and within the summer season). The proposed approach further
ADEV 96.9 79.1 1.4 1.0 1.4 1.0 allowed the investigation of correlations between the attributes. The
4 (Night-time) MEAN 55.2 58.8 5.6 5.7 1.6 1.6 methodology was applied to more than 3000 MODIS images acquired
MAX 315.0 188.0 12.3 9.1 8.3 5.1 from May until September from years 2000 to 2009 for the Greater
ADEV 45.6 33.4 1.0 0.8 1.0 0.7
Athens Area in Greece.
 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090 3089

covered by sparse low vegetation and bare soil and exhibited faster
heating rate than urban areas (negative heat island). On the other
hand, the city center of Athens was characterized by a strong
UHI phenomenon that was observed later in the day and mostly at
night-time. The daytime hot-spots exhibited similar behavior,
gradually increasing their temperature during the summer season,
reaching their maxima in mid-July. Their thermal intensities were of
9–10 °C. The areal extent was found to be highly correlated to the
thermal intensity. During night-time, Athens center developed a
typical UHI spatially coinciding with the dense urban fabric. The
maximum LST peaked (on average) at the end of July, showing a two-
week time lag compared to the daytime surface patterns.
The main innovation of this research is that the thermal hot-spots
were extracted and treated as objects. This allowed for the calculation
of several features related to the hot spots (e.g., the area extent, the
maximum, mean and minimum LST) from the original LST maps.
Other recent studies (Rajasekar & Weng, 2009; Streutker, 2003) had
dealt with modeled approximations (e.g. fast Fourier transformation;
FFT) of the temperature patterns. For instance, Streutker (2003)
approximated the area extent of the UHIs by calculating the
longitudinal and latitudinal extent assuming an ellipsoid footprint.
Additionally, our method was used to extract the attributes and to
present them as a series of functions rather than a series of images, as
suggested by Rajasekar and Weng (2009). The thermal features
retained their original values circumventing modeling and smoothing
out the maximum values of the original data. The proposed
methodology can be considered appropriate for both coastal cities,
where the discontinuity of sea-land interface would cause complica-
tions in the modeling approach, as well as inland cities. The
methodology can be customized to cover other geographical areas.
Nevertheless, sufficient knowledge of a city is required to efficiently
define the areas of interest as well as the reference rural/suburban
area. Information on the city's land cover and topography is also
important for the interpretation of the patterns. However, once the
area of interest is set, the algorithm processes thousands of LST
images in a few seconds without any manual interference. As com-
parison of thermal patterns across different cities is currently
hampered by the lack of an appropriate methodology to extract the
patterns and characterize them, future research should examine the
application of the method to different cities and comparison of their
UHIs behavior.

Acknowledgments

Special thanks go to Ms. Maria Michelaraki from the Hellenic

National Meteorological Service for her contribution regarding the
causes of the thermal patterns of Athens. The authors are also thankful
to Mr. Themistocles Herekakis (NOA/ISARS) for producing the plots.
The MODIS data used for this paper were procured and pre-processed
within the framework of Urban Heat Islands and Urban Thermogra-
phy project (www.urbanheatisland.info, 21913/08/I-LG; European
Space Agency). The authors wish to thank the anonymous reviewers
for their constructive comments that improved substantially the
manuscript.

References

Aniello, C., Morgan, K., Busbey, A., & Newland, L. (1995). Mapping micro Urban Heat
Islands using Landsat TM and a GIS. Comparative Geoscience, 21, 965–969.
Balling, R. C., & Brazel, S. W. (1988). High-resolution surface-temperature patterns in a
Fig. 7. Correlation between the thermal intensity and the surface extent of the daytime complex urban terrain. Photogrammetric Engineering & Remote Sensing, 54,
hot-spot of (a) Megara, (b) Elefsina-Aspropyrgos and (c) Mesogeia. 1289–1293.
Carnahan, W. H., & Larson, R. C. (1990). An analysis of an urban heat sink. Remote
The results show that in MODIS daytime acquisitions three areas Sensing of Environment, 33, 65–71.
Caselles, V., Lopez Garcia, M. J., Melia, J., & Perez Cueva, A. J. (1991). Analysis of the heat-
consistently appeared warmer than the city center. These areas are island effect of the city of Valencia, Spain through air temperature transets and
Megara, Elefsina-Aspropyrgos and Mesogeia which were mainly NOAA satellite data. Theoretical and Applied Climatology, 43, 195–203.
3090 I. Keramitsoglou et al. / Remote Sensing of Environment 115 (2011) 3080–3090
Dandou, A., Tombrou, M., & Soulakellis, N. (2009). The influence of the city of Athens
on the evolution of the sea-breeze front. Boundary-Layer Meteorology, 131,
35–51.
Dousset, B., & Gourmelon, F. (2003). Satellite multi-sensor data analysis of urban
surface temperatures and land cover. ISPRS Journal of Photogrammetry and Remote
Sensing, 58, 43–54.
Founda, D., & Giannakopoulos, C. (2009). The exceptionally hot summer of 2007 in
Athens, Greece—A typical summer in the future climate? Global and Planetary
Change, 67, 227–236.
Gallo, K. P., McNab, A. L., Karl, T. R., Brown, J. F., Hood, J. J., & Tarpley, J. D. (1993). The use
of a vegetation index for assessment of the Urban Heat Island effect. International
Journal of Remote Sensing, 14, 2223–2230.
Gallo, K. P., McNab, A. L., Karl, T. R., Brown, J. F., Hood, J. J., & Tarpley, J. D. (1993). The use
of NOAA AVHRR data for assessment of the Urban Heat Island effect. Journal of
Applied Meteorology, 32, 899–908.
Gallo, K. P., & Owen, T. W. (1998). Assessment of Urban Heat Island: A multisensor
perspective for the Dallas-Ft. Worth, USA region. Geocarto International, 13, 35–41.
Hung, T., Uchihama, D., Ochi, S., & Yasuoka, Y. (2006). Assessment with satellite data of
the Urban Heat Island effects in Asian mega cities. International Journal of Applied
Earth Observation and Geoinformation, 8, 34–48.
Imhoff, M. L., Zhang, P., Wolfe, R. E., & Bounoua, L. (2010). Remote sensing of the Urban
Heat Island effect across biomes in the continental USA. Remote Sensing of
Environment, 114, 504–513.
Jiménez-Muñoz, J. C., & Sobrino, J. A. (2008). Split-window coefficients for land surface
temperature retrieval from low-resolution thermal infrared sensors. IEEE Geoscience
and Remote Sensing Letters, 5, 806–809.
Kidder, S. Q., & Wu, H. T. (1987). A multispectral study of the St. Louis area under snow-
covered conditions using NOAA-7 AVHRR data. Remote Sensing of Environment, 22,
159–172.
Kim, H. H. (1992). Urban Heat Island. International Journal of Remote Sensing, 13,
2319–2336.
Lagouarde, J. -P., & Irvine, M. (2008). Directional anisotropy in thermal infrared
measurements over Toulouse city center during the CAPITOUL measurement
campaigns: First results. Meteorol Atmos Phys., 102, 173–185.
Lee, H. -Y. (1993). An application of NOAA AVHRR thermal data to the study of Urban
Heat Islands. Atmospheric Environment, 27B, 1–13.
Livada, I., Santamouris, M., Niachou, K., Papanikolaou, N., & Mihalakakou, G. (2002).
Determination of places in the great Athens area where the heat island effect is
observed. Theoretical and Applied Climatology, 71, 219–230.
Lo, C. P., Quattrochi, D. A., & Luvall, J. C. (1997). Application of high resolution thermal
infrared remote sensing and GIS to assess the Urban Heat Island effect. International
Journal of Remote Sensing, 18, 287–304.
Nichol, J. E. (1994). A GIS based approach to microclimate monitoring in Singapore's
high rise housing estates. Photogrammetric Engineering and Remote Sensing, 60,
1225–1232.
Nichol, J. E. (1996). Analysis of the urban thermal environment with LANDSAT data.
Environment and Planning B: Planning and Design, 23, 733–747.
Nichol, J. E. (1998). Visualization of urban surface temperatures derived from satellite
images. International Journal of Remote Sensing, 19, 1639–1649.
Nichol, J., & Wong, W. S. (2005). Modeling urban environmental quality in a tropical
city. Landscape and Urban Planning, 73, 49–58.
Oke, T. R. (1982). The energetic basis of the Urban Heat Island. Quarterly Journal of the
Royal Meteorological Society, 108, 1–24.
Oke, T. R. (1987). Boundary layer climates. Cambridge: University Press.
Owen, T. W., Carlson, T. N., & Gillies, R. R. (1998). An assessment of satellite remotely
sensed land cover parameters in quantitatively describing the climatic effect of
urbanization. International Journal of Remote Sensing, 19, 1663–1681.
Pal, N. R., & Pal, S. K. (1993). A review on image segmentation techniques. Pattern
Recognition, 26, 1277–1294.
Pu, R., Gong, P., Michishita, R., & Sasagawa, T. (2006). Assessment of multi-resolution
and multi-sensor data for urban surface temperature retrieval. Remote Sensing of
Environment: Thermal Remote Sensing of Urban Areas, 104, 211–225.
Rajasekar, U., & Weng, Q. (2009). Urban Heat Island monitoring and analysis by data
mining of MODIS imageries. ISPRS Journal of Photogrammetry and Remote Sensing,
64, 86–96.
Roth, M., Oke, T. R., & Emery, W. J. (1989). Satellite-derived Urban Heat Island from
three coastal cities and the utilization of such data in urban climatology.
International Journal of Remote Sensing, 10, 1699–1720.
Santamouris, M., Paraponiaris, K., & Mihalakakou, G. (2007). Estimating the ecological
footprint of the heat island effect over Athens, Greece. Climatic Change, 80,
265–276.
Sobrino, J. A., Jiménez-Muñoz, J. C., Sòria, G., Romaguera, M., Guanter, L., Moreno, J., et al.
(2008). Land surface emissivity retrieval from different VNIR and TIR sensors. IEEE
Transactions on Geoscience and Remote Sensing, 46, 316–327.
Sobrino, J. A., Kharraz, E. J., & Li, Z. -L. (2003). Surface temperature and water vapour
retrieval from MODIS data. International Journal of Remote Sensing, 20, 5161–5182.
Sobrino, J. A., Raissouni, N., & Li, Z. -L. (2001). A comparative study of land surface
emissivity retrieval from NOAA data. Remote Sensing of Environment, 75, 256–266.
Sobrino, J. A., Raissouni, N., Simarro, J., Nerry, F., & Petitcolin, F. (1999). Atmospheric
water vapor content over land surfaces derived from the AVHRR data: Application
to the Iberian Peninsula. IEEE Transactions and Geoscience and Remote Sensing, 37,
1425–1434.
Stathopoulou, M., & Cartalis, C. (2007). Daytime Urban Heat Islands from Landsat ETM
+ and Corine land cover data: An application to major cities in Greece. Solar Energy,
81, 358–368.
Stathopoulou, M., & Cartalis, C. (2007). Use of satellite remote sensing in support of
Urban Heat Island studies. Advances in Building Energy Research, 1, 203–212.
Stathopoulou, M., & Cartalis, C. (2009). Downscaling AVHRR land surface temperatures
for improved surface Urban Heat Island intensity estimation. Remote Sensing of
Environment, 113, 2592–2605.
Stathopoulou, M., Cartalis, C., & Keramitsoglou, I. (2004). Mapping micro-urban heat
islands using NOAA/AVHRR images and CORINE Land Cover: an application to
coastal cities of Greece. International Journal of Remote Sensing, 25, 2301–2316.
Streutker, D. R. (2002). A remote sensing study of the Urban Heat Island of Houston,
Texas. International Journal of Remote Sensing, 23, 2595–2608.
Streutker, D. R. (2003). Satellite-measured growth of the Urban Heat Island of Houston,
Texas. Remote Sensing of Environment, 85, 282–289.
Voogt, J. A., & Oke, T. R. (2003). Thermal remote sensing of urban climate. Remote
Sensing of Environment, 86, 370–384.
Weng, Q. (2001). A remote sensing-GIS evaluation of urban expansion and its impact on
surface temperature in the Zhujiang Delta, China. International Journal of Remote
Sensing, 22, 1999–2014.
Weng, Q. (2003). Fractal analysis of satellite-detected Urban Heat Island effect.
Photogrammetric Engineering and Remote Sensing, 69, 555–566.
Weng, Q. (2009). Thermal infrared remote sensing for urban climate and environ-
mental studies: Methods, applications, and trends (Review article). ISPRS Journal of
Photogrammetry and Remote Sensing, 64, 335–344.
Weng, Q., Liu, H., Liang, B., & Lu, D. (2008). The spatial variations of urban land surface
temperatures: pertinent factors, zoning effect, and seasonal variability. IEEE Journal
of Selected Topics in Applied Earth Observations & Remote Sensing, 1(2), 154–166.
Weng, Q., Lu, D., & Liang, B. (2006). Urban surface biophysical descriptors and land
surface temperature variations. Photogrammetric Engineering & Remote Sensing, 72,
1275–1286.
Weng, Q., Lu, D., & Schubring, J. (2004). Estimation of land surface temperature
vegetation abundance relationship for Urban Heat Island studies. Remote sensing of
Environment, 89, 467–483.
Weng, Q., & Quattrochi, D. A. (2006). Thermal remote sensing of urban areas: An
introduction to the special issue (Editorial). Remote Sensing of Environment, 104,
119–122.
WIST web site Created by WIST, Version WIST-10.30.GA-1. https://wist.echo.nasa.gov/~wist/
api/imswelcome/ (last accessed on 12/22/2010)
Xiao, R., Weng, Q., Ouyang, Z., Li, W., Schienke, E. W., & Zhang, W. (2008). Land surface
temperature variation and major factors in Beijing, China. Photogrammetric
Engineering & Remote Sensing, 74, 451–461.